fix: truncate oversized sections and clear DB on fresh ingest

- Truncate content to 3800 chars before embedding (embeddinggemma has
  a 2048-token context, ~4K char limit). Eliminates all 73 Ollama 500
  errors from oversized plan sections.
- Clear brain_memories DB table when --fresh is used, keeping DB rows
  in sync with Qdrant vectors.

Co-Authored-By: Virgil <virgil@lethean.io>

Console/Commands/BrainIngestCommand.php

@@ -133,11 +133,19 @@ class BrainIngestCommand extends Command
             }
 
             try {
+                $text = $section['heading']."\n\n".$section['content'];
+
+                // embeddinggemma has a 2048-token context (~4K chars).
+                // Truncate oversized sections to avoid Ollama 500 errors.
+                if (strlen($text) > 3800) {
+                    $text = mb_substr($text, 0, 3800).'…';
+                }
+
                 $brain->remember([
                     'workspace_id' => $workspaceId,
                     'agent_id' => $agentId,
                     'type' => $type,
-                    'content' => $section['heading']."\n\n".$section['content'],
+                    'content' => $text,
                     'tags' => $tags,
                     'project' => $project,
                     'confidence' => $this->confidenceForSource($source),
@@ -400,9 +408,13 @@ class BrainIngestCommand extends Command
         $prop = $reflection->getProperty('collection');
         $collection = $prop->getValue($brain);
 
+        // Clear Qdrant collection.
         \Illuminate\Support\Facades\Http::withoutVerifying()
             ->timeout(10)
             ->delete("{$qdrantUrl}/collections/{$collection}");
+
+        // Truncate the DB table so rows stay in sync with Qdrant.
+        \Core\Mod\Agentic\Models\BrainMemory::query()->forceDelete();
     }
 
     private function expandHome(string $path): string